【摘要】 为克服钢铁行业长期电力高噪声和高负荷带来的非线性和非平稳的预测挑战，提出完全集成经验模态分解(CEEMDAN)、长短期记忆网络(LSTM)与序列到序列结构(Seq2Seq)的CLSG组合预测模型。首先，基于CEEMDAN分解原始负荷序列，提取多尺度模态分量；其次，采用LSTM-Seq2Seq模型捕捉负荷数据的时序依赖关系与序列演化特征，通过网格搜索进行关键参数寻优；最后，以云南曲靖钢铁行业电力负荷数据开展实验验证分析和对比分析。研究结果表明：CLSG模型的平均绝对误差在0.1以内，均方根误差在0.15以内，平均绝对百分比误差在0.2以内，相较于TBA、CRSG、CGSG、MCLS模型，CLSG模型的误差指标值均最小，具有更高的精度与稳定性。研究结果可为钢铁行业电力负荷精准预测与高效管理提供新方法。更多还原

【Abstract】 In order to overcome the challenges of nonlinear and non-stationary prediction caused by the high noise and high load in the steel industry’s long-term electricity usage, a fully integrated prediction model named CLSG is proposed, which combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN),Long Short-Term Memory Network(LSTM),and Sequence-to-Sequence(Seq2Seq) architecture.Firstly, the original load sequence is decomposed based on CEEMDAN,extracting multi-scale modal components.Secondly, the LSTM-Seq2Seq model is employed to capture the temporal dependencies and sequence evolution characteristics of the load data, with key parameters optimized using grid search.Finally, experimental verification and comparative analysis are conducted based on the power load data from the steel industry in Qujing, Yunnan.The results indicate that the CLSG model achieves an average absolute error of less than 0.1,a root mean square error of less than 0.15,and an average absolute percentage error of less than 0.2.Compared to the TBA,CRSG,CGSG,and MCLS models, the CLSG model has the smallest error metrics, demonstrating higher accuracy and stability.The findings provide a new approach for precise prediction and efficient management of electricity loads in the steel industry.更多还原